OpenAI Gives ChatGPT Direct Access to Your iPhone Messages Inbox

Aug 210 views

Key takeaways

ChatGPT's new Apple Messages plug-in can draft, send, delete, and search iMessages on a user's behalf.
OpenAI says the plug-in runs locally and does not index all messages, but full data details remain unclear.
OpenAI warns users against enabling persistent approval, which removes manual review before messages are sent.

OpenAI has released an Apple Messages plug-in for ChatGPT, giving the chatbot the ability to read, draft, send, and delete messages directly from a user's iMessage inbox. The plug-in also integrates with Codex and ChatGPT Work, positioning it for professional use cases alongside personal messaging. A promotional video released alongside the launch shows a user prompting ChatGPT to suggest follow-up replies based on messages received the previous day — a concrete example of how the feature is intended to function in everyday life.

Beyond composing replies, users can instruct ChatGPT to search through old message threads, delete specific conversations, or send messages entirely on their behalf. The scope of that access is notably broad: the chatbot can act as an autonomous agent inside one of the most personal communication channels on an iPhone. OpenAI has positioned this as a productivity tool, but the extent of what data the model can access during a session has not been spelled out in granular detail publicly.

Privacy concerns surfaced almost immediately after the announcement. OpenAI told Bloomberg that the plug-in runs locally on a user's machine and does not build a persistent index of message history. What exactly that means in practice — what is cached, what is processed, and for how long — remains unclear. TechCrunch confirmed it has reached out to OpenAI for additional clarification on those specifics.

OpenAI itself appears cautious about the autonomous-sending aspect of the feature. The company actively discourages users from enabling persistent approval mode, warning that doing so eliminates the final manual review step before ChatGPT transmits a message under the user's name. That warning signals OpenAI is aware of the reputational and practical risks if the chatbot sends an unintended or incorrect message to a contact.

The plug-in arrives as OpenAI continues to expand ChatGPT's integration with operating-system-level features, moving beyond the chat window and into core device functions. Apple's own AI efforts under the Apple Intelligence brand have proceeded at a slower pace than many expected, which may make third-party integrations like this one more prominent in the near term among users looking for smarter messaging assistance.

The bigger picture

The Apple Messages plug-in is a meaningful step in OpenAI's strategy to embed ChatGPT into the fabric of daily device use rather than keep it siloed as a standalone chat product. By reaching into iMessage — arguably the stickiest communication app for iPhone users in the United States — OpenAI is competing on the same turf Apple Intelligence is supposed to occupy. If ChatGPT proves more capable or responsive than Apple's native AI features, it could erode Apple's argument that deep system integration gives its own AI an inherent advantage.

The privacy framing here deserves scrutiny beyond the 'runs locally' assurance. Running locally does not automatically mean data is not transmitted anywhere; it describes where computation happens, not necessarily what leaves the device. Regulators in the EU, where data minimization requirements under GDPR are strict, will likely want specifics. Rivals like Google, which is building similar agentic capabilities into Gemini for Android, will be watching how OpenAI handles any backlash — either to learn from it or to use it as a cautionary selling point for their own approach.

For everyday users, the more immediate question is trust calibration. OpenAI's own warning against persistent approval is essentially an admission that fully autonomous message-sending carries real risk. That tension — between convenience and control — will define how broadly this plug-in is adopted. If a high-profile mishap occurs where ChatGPT sends an embarrassing or incorrect message on someone's behalf, the resulting press could set back agentic AI adoption across the industry, not just at OpenAI.

LagPing's take

We're covering this at LagPing because an AI chatbot gaining send-level access to private messages is a genuinely significant threshold, not just another feature drop. Most prior ChatGPT integrations have been about information retrieval or content generation inside the app itself — this one reaches into a communication channel people use for everything from family check-ins to work negotiations. That shift in scope matters, and we think our readers deserve a clear-eyed look at both the functionality and the unanswered questions around it. We're also watching this closely because it sits at the intersection of Apple's own AI ambitions and OpenAI's platform expansion, a rivalry that is quietly shaping what our devices can do without us lifting a finger. We'll be following up as OpenAI responds to outstanding questions about data handling.

X / Twitter WhatsApp

View at TechCrunch AI

Find "ChatGPT" on Amazon

As an Amazon Associate, LagPing earns from qualifying purchases. Product links are affiliate links.